fangjet
// ai-assisted development

Using AI to build faster
without lowering the bar.

Tools change. Judgment doesn't.

claude-code · fangjet
>find why reminders fire an hour late after the DST change
Reading ReminderScheduler.kt, AlarmRepository.kt...
Cause: trigger time stored as local time, not UTC
>write a failing test for that first
✓ ReminderSchedulerTest: dstForwardShift (fails)
>now fix it
✓ 1 file changed, test passes
>
// how I use it
>_

Architecture exploration

Mapping unfamiliar codebases, finding the patterns, and planning refactors before any code changes.

>_

Test generation

Building unit tests for business logic quickly, then reviewing and hardening what comes out.

>_

Code review

Identifying the design decisions hiding inside a diff, not just the syntax.

>_

MCP integration

Built an MCP connector for Rolo CRM so AI agents can read contacts, log activities, and surface follow-ups through natural conversation instead of a UI.

>_

Documentation

First drafts of architecture docs and inline comments, kept honest by someone who knows the code.

>_

Build and release

Diagnosing CI failures, unblocking signing issues, and keeping the pipeline green.

AI tools can get you to a working first version fast. They can also generate bloated solutions, unnecessary complexity, and code that proves hard to maintain. The value isn't in the output. It's in knowing what to keep.